The project aims at developing improved techniques for data analysis and control of critical infrastructure safety, with the ultimate aim of:

• Get a better understanding of the actual safety state of the structures to efficiently manage the available resources with regard to: a) defining priorities among the repair, maintenance and improvement activities to undertake, and b) avoiding unnecessary actions.
• Get a better understanding of the actual behavior of structure with the aim of lengthening the service life and increasing the safety standards.
• Be able to anticipate possible failures and breakdow

CIMNE is working on the development and application of machine learning tools to build predicting models. They give an estimate of the performance of the structure in safety conditions, which can be compared with the actual measurements: if the discrepancy exceeds a certain threshold, the system issues a warning to dam safety responsible.

Some of the tools analyzed are neural networks, random forests, support vector machines, boosted regression trees and multi-adaptive regression splines (MARS). The preliminary results support the conclusion that these new models are more flexible and offer greater accuracy than conventional statistical methods.

Currently, the ability of these models to identify patterns of behavior is being assessed, in order to better understand dam behavior.

In parallel, these tools are being integrated into web applications, so that they are fully accessible by the authorized users.

     
Figure 1: Web application for building a random forest model. The user sets the model meta-parameters with the controls in the left column, and the results are displayed on the right. The app also features a brief explanation of the model description and links to useful resources.
 

Figure 2: Web application for building a random forest model. Example video.